白光阴道镜图像对比度较低, 不利于医生鉴别不同病变程度组织, 也不利于自动化宫颈癌筛查。 利用癌变组织富含血红蛋白成分及血红蛋白具有特征波段这一特性, 与传统高光谱空间扫描成像及分时获取不同波段多光谱成像方法相反, 利用快照式多光谱窄带成像来加速光谱图像获取过程, 提升不同病变程度组织之间灰度对比度同时, 降低后续图像分析处理算法难度, 实现对宫颈组织病变类型高帧率自动化分类。 首先, 使用微型快照式窄带多光谱摄像方法, 在血红蛋白的两个强吸收峰(415±10)和(525±10) nm、 一个反射带(620±10) nm和一个背景波段(450±10) nm共四个波段对宫颈组织进行快照式零时差获取四幅窄带光谱图像。 而后, 对所获取的光谱图像进行简单代数加减, 以生成突显病变组织的融合图像, 提高不同病变程度组织之间的对比度。 最后, 使用欧式距离分类算法, 对光谱融合图像中不同病变级别进行分类, 建立计算机辅助宫颈癌筛查方法。 创新点在于实现了高帧率计算机辅助光学病理诊断方法。 分别采用临床常规白光阴道镜及微型快照式窄带多光谱摄像对宫颈癌手术切下的新鲜组织进行彩色图像及光谱融合图像的高帧率采集, 并使用同一个欧式距离分类算法对两种图像进行自动分类, 分类结果都以组织病理诊断作为标准来计算正确率。 通过对比两种分类结果正确率来检验光谱融合图像相对于彩色图像是否提升对比度, 及其是否可以实现与组织病理诊断(金标准)结果一致的诊断。 欧式距离分类算法对光谱融合图像分类准确率接近100%, 远高于对白光阴道镜图像约50%的准确率。 多位临床医生对基于微型快照式多光谱摄像头光谱融合图像的计算机自动分类结果表示接受。 微型快照式窄带多光谱成像方法能有效提升光谱融合图像获取帧率及不同病变程度组织之间灰度对比度, 能有效快速地将宫颈组织划分为与组织病理诊断结果一致的病变类型。 由于诊断客观、 无创伤、 结果立等可得, 该方法将有助于实现落后地区宫颈癌筛查的普及以及图像导航下的宫颈癌精准治疗手术。

Indeed, the contrast of white-light colposcope image is rather low, which would not be ideal for cervical cancer screening. According to the fact that cancerous tissue has a rich composition of hemoglobin, which has multiple characteristic spectral bands. In contrary to traditional hyperspectral imaging or sequential multi-spectral imaging, which needs scanning in the either spatial or spectral domain, here, we propose the utilization ofminiature snapshot narrow-band imaging (SNBI) method to expedite the spectral image acquisition process and enhance the contrast between different tissues. The goal is to realize a fast computer-aided diagnosis method for early screening of cervical cancer. Firstly, we usedan SNBI technology to capture the images of cervix tissues at four characteristic bands of hemoglobin, namely two of its absorption peaks at (415±10) and (525±10) nm, one reflectance peak at (620±10) nm, and one background band at (450±10) nm. Secondly, we fused those spectral images via simple algebric operation to enhance the contrast between normal and abnormal tissues. Thirdly, Euclidean distance algorithm was applied to the fused image to classify the tissues into different lesion grades. This is the first computer aided optical pathological diagnosis method with a diagnosis rate of over 20 fps. Herein, white-light colposcope, and miniature SNBI video camera were usedto separately capture images of fresh cervical tissues that were surgically dissected within 10 minutes. The same Euclidean distance classification algorithm was applied to the images obtained by the white light colposcope, and to the spectrally fused image obtained by the SNBI video camera. The classification accuracy of the two imaging methods was calculated and compared, using the histopathologic diagnosis as a standard reference. Euclidean classification accuracy upon the spectral fused image acquired by the SNBI was approximately 100%, which is undoubtedly better than that of the color image acquired by the conventional colposcopy. Multiple experienced gynecologists also subjectively agreed with the computer-generated classification upon the fused image, and highly acknowledged its clinical value especially on challenging areas where multiple degreed lesion layered together. The proposed SNBI method could improve the acquisition frame rate and contrast of the spectrally fused image, and effectively classify the cervical tissue into pathological-diagnosis-consistent types of tissues. Due to its advantages of being objective, intact and instant, SNBI has excellent potential to enlarge the coverage of cervical screening population in a low-income district and to assistprecise treatment of cervical cancer under image guidance.